Thread | Brutkey

I provided Claude COde and Gemini with the same prompt to generate a requirement document for rewriting an existing hTML form with some JS/CSS and a bit of simple field dependencies (show/hide) and some average validations...Gemini failed completely (it did not even include the possible values of select nodes, if when instructed multiple times). Claude Code generated a very reasonable and comprehensive with all aspects...cookie to Claude AI ...
#AI #VideCoding #ClaudeCode #Gemini

Andreas 🌈

@zopyx@mastodon.world

The presented the specs of the form for a complete clean-room implementation to Gemini 2.5 and Claude Code...again, Gemini implemented only a part of the functionality and change change field labels unanticipated etc...completely useless.. Claude Code generated a fully functional UI in the first iteration but destroyed its own implementation while fixing or adjusting minor details with further prompting..in the end...a failed experiment...

#AI #VideCoding #ClaudeCode #Gemini

Andreas K
@yacc143@mastodon.social

@zopyx@mastodon.world Which gemini model, which tool.

You are comparing apples with non-apples here.

A coding aid, LLM based is the whole of the application, prompts, and LLM. If you have a problem with this, compare Sonnet-4 in Claude code versus Copilot. Same model, but even with copilot in agentic mode, the "vibe" (pun intended) is quite different.

And you are comparing two completely different products here. Claude Code costs roughly $200/m if you plan to use it on anything more than toy projects.

Andreas 🌈

@zopyx@mastodon.world

@yacc143@mastodon.social the Codex cli, gemini cli and claude cli are made form programming against OpenAi, Gemini Flash ans Claude Code…I don’t care about the underlaying models. I am just using these tools.